A Context Model Approach to Anaphora Resolution in Database-Oriented Discouse

概要

論文の詳細を見る
This paper proposes a Context Model approach to resolve Anaphora Problem in Database-Oriented Discourse. The essential idea is : for a limited speciflc domain, e. g., database, each object type can be regarded as the conveyer of the proceeding in discourse. Since the introductory sentence of a discourse always provides a means of introducing a new object with its type for further discussion, it is reasonable to build a Context Model (CM) as a premodel of discourse which consists of all expected attributes of that object. With the strong intuition, certain attributes belong to certain objects. Then expected context models can be used to unify with the attribute values given by the next sentences which may be related to the antecedent sentence by using anaphora references. Those unified sentences will give the answers of anaphora referents. To effectively realize such the idea, an Attribute-oriented Logical Form Language (ALFL) is also formulated as the meaning representation language for supporting the process of using CM, such as inferring new information, expecting presupposed objects' attributes from knowledge rules and obtaining the anaphora resolution through unification process.
社団法人人工知能学会の論文
1990-11-01